Approaches for Lossy Compression Using Machine Learning

1. A computing system comprising: one or more processors; and memory storing instructions that, when executed by the one or more processors, cause the system to perform: obtaining a media, the media including an image, audio, video, or combination thereof; receiving an input regarding one or more features and frames of the media to be maintained in or removed from the media, wherein the receiving of the input comprises receiving a selection of a first feature while a second feature remains unselected; inferring, using a machine learning model, a first criteria and a second criteria of a lossy compression technique, based on the received input, wherein the inferred first criteria is applied to any first features matching or having a same classification as the first feature and the inferred second criteria is applied to one or more second features having a different classification compared to the first feature, wherein the inferred first criteria or the inferred second criteria is based on one or more relative amounts or rates of movement of one or more objects across the one or more frames; and applying the inferred first criteria and the inferred second criteria of the lossy compression technique to the media, wherein the applying of the inferred first criteria and the inferred second criteria comprises applying the inferred first criteria to the any first features and applying the inferred second criteria to the one or more second features.

2. The computing system of claim 1, wherein the instructions further cause the system to perform: in response to applying the inferred first criteria and the inferred second criteria of the lossy compression technique, receiving an additional input; adjusting the inferred first criteria or the inferred second criteria based on the additional input; and reapplying the adjusted inferred first criteria or the adjusted inferred second criteria of the lossy compression technique to the media.

3. The computing system of claim 1, wherein the inferred first criteria or the inferred second criteria is applied to an action of an other feature or frame in the media, the action indicating whether to downsample, compress, keep, or remove the other feature or frame.

4. The computing system of claim 1, wherein the inferred first criteria or the inferred second criteria is determined based on a scale of a byte or a frame, the criteria indicating whether to downsample, compress, keep, or remove the byte or the frame.

5. The computing system of claim 4, wherein the inferred first criteria or the inferred second criteria further comprises, in response to determining to remove the byte or the frame, determine whether to purge the byte or the frame or to store the byte or the frame in a tiered storage.

6. The computing system of claim 1, wherein: the applying, using the machine learning model, the inferred first criteria and the inferred second criteria of the lossy compression technique to the media, comprises: determining, in each frame of the media, one or more bounding regions, each of the bounding regions enclosing an entity; across the frames, determining, using instance segmentation or semantic segmentation, bounding regions that enclose an entity that matches the classification of the first feature identified in the received input; and applying the inferred first criteria to the determined bounding regions.

7. The computing system of claim 6, wherein the determining, using semantic segmentation, bounding regions that enclose an entity having a same classification as the first feature identified in the received input comprises determining bounding regions that enclose an entity having a probability of having a same classification as the first feature satisfying a threshold.

8. The computing system of claim 6, wherein the bounding regions comprise bounding boxes, and the machine learning model is trained based on a bounding box loss and a classification loss.

9. The computing system of claim 1, wherein the machine learning model is trained to infer the first criteria and the inferred second criteria based on different lighting conditions.

10. The computing system of claim 1, wherein the input comprises tags of a portion of the media; and the applying of the inferred first criteria and the inferred second criteria comprises applying the inferred first criteria or the inferred second criteria to one or more untagged portions of the media.

11. The computing system of claim 10, wherein the inferred first criteria or the inferred second criteria is determined based on a frequency at which frames of the media are tagged.

12. The computing system of claim 10, wherein the inferred first criteria or the inferred second criteria is determined based on a spatial pattern, arrangement, relationship, or configuration among tagged features of the media.

13. The computing system of claim 1, wherein the inferred first criteria or the inferred second criteria is determined based on a sequence of events or movements among frames of the media.

14. The computing system of claim 1, wherein the inferred first criteria or the inferred second criteria is determined based on a perceived spatial relationship between two entities in the media.

15. The computing system of claim 1, wherein the inferred first criteria or the inferred second criteria is determined based on a perceived size of an entity in the media.

16. The computing system of claim 1, wherein the instructions that, when performed by the one or more processors, further cause the computing system to perform: generating a compressed version of the media by applying the lossy compression technique, the lossy compression technique resulting in compressed features; storing the compressed version and the compressed features of the media within a storage according to a storage criteria, wherein the storage criteria comprises storing portions of the compressed version or the compressed features that have higher information content or higher information value within a first location and portions of the compressed version or the compressed features having lower information content or lower information value within a second location, wherein the first location is configured to retrieve the compressed version of the unstructured information faster compared to the retrieving of the portions of the compressed features from the second location.

17. The computing system of claim 1, wherein the inferred first criteria compresses the any first features at a first compression rate and the inferred second criteria compresses the one or more second features at a second compression rate.

18. The computing system of claim 17, wherein the first compression rate is higher than the second compression rate.

19. The computing system of claim 1, wherein the inferred first criteria and the inferred second criteria are applied to the any first features and the one or more second features within a same image frame.

20. A computer-implemented method, wherein the method is performed using one or more processors, the method comprising: obtaining a media, the media including an image, audio, video, or combination thereof; receiving an input regarding one or more features and frames of the media to be maintained in or removed from the media, wherein the receiving of the input comprises receiving a selection of a first feature while a second feature remains unselected; inferring, using a machine learning model, a first criteria and a second criteria of a lossy compression technique, based on the received input, wherein the inferred first criteria is applied to any first features matching or having a same classification as the first feature and the inferred second criteria is applied to one or more second features having a different classification compared to the first feature, wherein the inferred first criteria or the inferred second criteria is based on one or more relative amounts or rates of movement of one or more objects across the one or more frames; and applying the inferred first criteria and the inferred second criteria of the lossy compression technique to the media, wherein the applying of the inferred first criteria and the inferred second criteria comprises applying the inferred first criteria to the any first features and applying the inferred second criteria to the one or more second features.